IBM's Watson gets chatty to act as a sounding board
While researchers around the world are making gradual gains in the monumental task of developing artificial intelligences that can creatively solve problems or produce art, IBM's Watson supercomputer has now learned how to help people get more creative. Six student teams at Georgia Tech trained Watson to chat with them about the many systems from nature that we could mimic in solving big problems such as long-term space travel and more efficient desalination processes.
The students fed Watson several hundred articles from the Biologue biology repository, then each team asked the AI 200 natural language questions – amassing a total of 1,200 question and answer pairs. Watson delivered its detailed, multi-referenced answers in a fraction of a second. To a question about how to develop better solar cells for long-term space travel, Watson suggested looking at how plants in harsh climates regulate their internal temperature using fibrous insulation. When asked to offer a better process for desalination of sea water for drinking purposes, the AI brought up multiple examples from the animal world – one of which was the glands that seagulls use to filter out salt from seawater.
This "GT-Watson Plus" technique, as the students named it, also offers a treetop-style visualization that displays relevant concepts in "leaves" that are sized according to their relevance or importance and linked to additional questions. And Watson can suggest alternative ways to phrase a question that might yield better results.
The students expect that this new technique will make Watson an intelligent sounding board in a variety of information-rich fields such as engineering, computing, architecture, science, and economics. It can effectively chat with students and professionals like a colleague or advisor, with its biggest boon being that it can help us filter and make sense of the dizzying amount of potentially-relevant research across, within, and between different fields.
What Watson is learning now could well be a glimpse of the future of digital personal assistants like Siri and web search engines like Google, which already have begun to offer context-sensitive responses to a variety of what, how, when, and why queries. Our access to information threatens to soon outstrip our ability to consume, understand, and interpret it, and AIs like Watson could be critical to keeping this shift in check as well as to expanding our minds by showing us the links between seemingly disparate ideas.
Georgia Tech professor Ashok Goel, who led the study, plans to next try applying Watson's nous to online learning and healthcare, among other areas. The famous supercomputer has previously mastered debating, Jeopardy, recipe making, customer service, and more.
The research was presented this week at the Association for the Advancement of Artificial Intelligence (AAAI) 2015 Fall Symposium on Cognitive Assistance in Government.
Source: Georgia Tech